> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orvado.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# GPT Image 2.5

> Output sizes, quality levels, image count and token pricing for GPT Image 2.5 Flare and Sunburst

GPT Image 2.5 has two models with the same price and the same parameters:

* `gpt-image-2.5-flare` for everyday images and drafts.
* `gpt-image-2.5-sunburst` for detailed images and precise text.

Both use `POST /v1/images/generations`, like any other [image model](https://docs.orvado.dev/images.md). This page lists what each parameter actually does, measured on our production API.

## Output size

The `size` you send sets the shape and, from 2K up, the pixel size of the image. Checked on 12 November 2026:

| `size` you send | Image you get | Output tokens at `medium` |
| - | - | -: |
| `1024x1024` | 1254 × 1254 | 1,650 |
| `1536x1024` | 1536 × 1024 | about 1,600 |
| `1024x1536` | 1024 × 1536 | about 1,600 |
| `2048x2048` | 2048 × 2048 | 4,224 |
| `3840x2160` (16:9) | 3840 × 2160 | 8,415 |

Larger images use more output tokens and cost more: a 2K `medium` image is about $0.019, a 4K `medium` image about $0.038. `background: "transparent"` is not supported; the image comes back as a regular PNG.

Always read the real width and height from the returned file instead of assuming the `size` you sent.

## Quality

`quality` changes both the detail and the price.

| `quality` | Output tokens per image | Price per image |
| - | -: | -: |
| `low` | about 425 | about \$0.002 |
| `medium` or `auto` (default) | about 1,650 | about \$0.0074 |
| `high` | about 6,500 | about \$0.029 |

`high` is the most detailed level. Requests that send `xhigh` or `max` are processed as `high`.

The token counts are typical values. The exact count for each image is in the `usage` field of the response, and you are charged for that count.

## Number of images

`n` can be **1 to 4**. A request with `n` above 4 returns a `422` error and is not charged. For more images, send several requests in parallel.

Each image in the response is charged by its own tokens: four `medium` images cost about 4 × \$0.0074.

## Pricing

GPT Image 2.5 is billed by tokens, like the official API, at 15% of the official rate:

| Tokens | Price per 1M | Official price per 1M |
| - | -: | -: |
| Text input (your prompt) | \$0.75 | \$5 |
| Cached text input | \$0.1875 | \$1.25 |
| Image input | \$1.20 | \$8 |
| Image output | \$4.50 | \$30 |

A prompt is usually under 100 tokens, so the price per image is almost entirely image output: output tokens × \$4.50 / 1,000,000.

When a request starts, Orvado holds \$0.10 per image on your balance. After the image is ready, the hold is replaced by the actual token price. A request that fails returns the hold in full.

## Example

```bash theme={null}
curl https://orvado.dev/v1/images/generations \
  --max-time 600 \
  -H "Authorization: Bearer $ORVADO_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.5-sunburst",
    "prompt": "poster with the text ORVADO 2026 in bold black letters",
    "size": "1536x1024",
    "quality": "high",
    "n": 2
  }'
```

This returns two 1536 × 1024 images for about \$0.056 in total. The `usage.output_tokens` field shows the tokens charged:

```json theme={null}
{
  "created": 1792700000,
  "data": [{ "url": "https://..." }, { "url": "https://..." }],
  "usage": {
    "input_tokens": 14,
    "output_tokens": 12480,
    "output_tokens_details": { "image_tokens": 12480 }
  }
}
```

Here 12,480 × $4.50 / 1,000,000 = $0.0562.

## Edit an image

GPT Image 2.5 edits use `POST /v1/images/edits` with input images as multipart `image[]` files or a JSON `image_url`. An edit is billed by tokens like a generation, plus the input images as image input tokens at $1.20 per 1M. A 1024 × 1024 photo is about 1,800 input tokens, so a `medium` edit costs about $0.0096. A `mask` is not supported. See [Edit an image](https://docs.orvado.dev/images.md#edit-an-image).

## Response time

GPT Image 2.5 takes about 25 to 60 seconds per request. Set your client timeout to at least 180 seconds; 600 seconds is safest. See [timeouts](https://docs.orvado.dev/images.md#response-time-and-timeouts).

> [!NOTE]
>
> Measured on 22 and 23 October 2026. If the model starts returning larger images, this page will be updated.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.